Rescue of Hypoxia-inducible Factor-1α-deficient Tumor Growth by Wild-Type Cells Is Independent of Vascular Endothelial Growth Factor

INTRODUCTION

Solid tumor growth depends crucially on neoangiogenesis, a process by which new blood vessels sprout from preexisting ones. However, the newly formed irregular vasculature is incapable of fully meeting the tumor tissue’s demands for oxygen and nutrients, leading to areas of acidosis and hypoxia. Therefore, neoplastic tissue exhibits lower oxygenation levels than the corresponding healthy tissue (reviewed in Ref. 1 ). This is of major importance in oncology because tumor hypoxia is associated with resistance to chemo- and radiotherapies, with neoplastic malignant progression, and with poor prognosis (2, 3, 4) . Thus, the development of drugs that are selectively activated under hypoxic conditions has opened new perspectives in cancer treatment in recent years (5 , 6) .

One of the major consequences of tissue hypoxia/anoxia is the protein stabilization of HIF-1α, 3 which belongs to a family of constitutively expressed bHLH proteins containing a Per/Arnt/Sim (PAS) motif (7) . Although other members of the HIF family have been cloned (8 , 9) , only HIF-1α is ubiquitously expressed. HIF-1α is constantly synthesized and degraded under normoxic conditions to ensure an extremely rapid response to hypoxia in vitro and in vivo (10 , 11) . This response is given through HIF-1α stabilization, translocation into the nucleus, and dimerization with its partner, ARNT/HIF-1β, forming a heterodimer called HIF-1 (12 , 13) . This heterodimer binds to the HREs present in various genes that promote the physiological response to hypoxia, including the up-regulation of VEGF, the most potent angiogenic molecule known, as well as glycolytic enzymes and glucose transporters (14) . The up-regulation of these genes contributes to tumor growth by enhancing the supply of glucose and its metabolism in neoplastic cells (15) , whereas VEGF plays a central role because of its angiogenic properties (16) .

The loss of tumor suppressor genes such as PTEN (17) and pVHL (18) or oncogenes such as v-src (19) has also been shown to increase HIF-1α expression. pVHL is directly involved in the degradation pathway of HIF-1α. In oxygenated cells, HIF-1α is hydroxylated within its oxygen-dependent degradation domain (20 , 21) , which enables the binding of HIF-1α to the β-domain of pVHL (22) . The loss of pVHL leads to the stabilization of HIF-1α under normoxic conditions and facilitates formation of hypervascularized tumors (23) .

The importance of HIF-1α for human tumor development has recently been documented. HIF-1α is overexpressed in the majority of human cancers (24) , with expression levels correlating with malignancy and negative survival prognosis (25 , 26) . Despite all this evidence, the exact role of HIF-1α in tumor biology is still controversial because the results from various tumor models are contradictory. The absence of HIF-1α in teratocarcinomas (27) or the disruption of the HIF-1α transcriptional coactivator p300 in human tumor cells reduced tumor growth (28) , but in one report, HIF-1α-deficient (HIF-1α^−/−) tumors showed increased growth compared with HIF-1α wild-type (HIF-1α^+/+) tumors because of an increased apoptosis rate in the HIF-1α^+/+ cells (29) .

During the past few years, several studies were performed using the hypoxic environment as an activator of gene expression driven by several concatamerized HREs (30 , 31) ; thus, enzyme-prodrug systems for hypoxia-targeted gene therapy are being developed. Furthermore, design of a gene therapy using transfer of antisense HIF-1α has been shown to enhance therapeutic efficacy when combined with cancer immunotherapy (32) . However, before HIF-1α can be used as a target for tumor treatment, the precise roles of HIF-1α-positive and -negative cells in a heterogeneous population must be determined. Therefore, we analyzed HIF-1α^+/+ and HIF-1α^−/− pure and mixed tumors in nude mice.

MATERIALS AND METHODS

Cell Culture

HM-1 ES cells (a generous gift from D. W. Melton, Edinburgh, United Kingdom) were cultivated on gelatin-coated Petri dishes as described previously (33) . EBs were formed using the hanging drop method as described elsewhere (34) . Oxygen tensions in the incubators were either 140 mm Hg (20% O₂, v/v; normoxia) or 7 mm Hg (1% O₂, v/v; hypoxia). Where indicated, 260 μm DFX or 100 μm CoCl₂ (both from Sigma Chemical Co, St. Louis, MO) were added.

Construction of the HIF-1α Targeting Vector and Generation of a HIF-1α Double-deficient Cell Line

A targeting vector (pSP73SGL) was constructed by ligating the 1.9-kb EcoRV-EcoRI (short arm) and a 7.5-kb PstI fragment (long arm) derived from the mouse HIF-1α genomic clone λH1 (35) into the EcoRV-EcoRI and PstI sites, respectively, of pSP73 (Promega, Madison, WI). A 6.7-kb SalI fragment derived from pGT1.8IRESβgeo (a gift from P. Mountford) was inserted into the SmaI site between the two arms, producing pSP73SGL. This promotorless fragment contained an En-2 splice acceptor site, an IRES and a fusion between the lacZ and neo genes (36) . pSP73SGL was linearized with XhoI and electroporated into HM-1 wild-type ES cells. Cells were selected for resistance to G418 (250 μg/ml; Calbiochem, La Jolla, CA). HIF-1α^+/− heterozygous clone 33 was then subjected to increasing concentrations of G418 to generate HIF-1α double-deficient clones, named C, L, and M.

Growth Curves

ES Cells.

ES cells were seeded at a density of 10⁴ cells/well (day 0) in 12-well plates (Fisher Scientific, Genolier, Switzerland) and cultivated for a maximum of 6 days in the presence or absence of LIF under normoxic or hypoxic conditions. Triplicate samples from HIF-1α^+/+ and HIF-1α^−/− cells were used, and medium was changed every second day to avoid nutrient exhaustion. Wells were rinsed twice with PBS and incubated for 5 min in 0.1% crystal violet in H₂O (Merck, Darmstadt, Germany). Crystals were dissolved in 1% SDS in H₂O, and the absorbance was measured at 550 nm.

EBs.

The size of at least 50 EBs was measured on days 2, 4, and 6 of their development, using an inverted microscope (Model IMT-2; Olympus, Hamburg, Germany) supplied with a scaled eyepiece. The exact EB diameter was calculated in mm.

Tumor Growth in Nude Mice

Female BALB/c nu/nu mice were obtained from IFFA Credo (L’Arbresle, France). The mice were kept and treated according to institutional guidelines approved by the Kantonales Veterinäramt Zürich. HIF-1α^+/+ and HIF-1α^−/− (clone L) ES cells were mixed in vitro at a 1:10 or 1:100 ratio before injection. Mice were inoculated s.c. with 5 × 10⁶ HIF-1α^+/+ or HIF-1α^−/− (clone L or M) cells or with 1:10 and 1:100 mixtures (n = 10 mice/group). Tumor size was measured with a caliper once or twice per week, and the volume was estimated according to the formula: volume (cm 3 ) = 1/2(L × W²), where L and W are the length (cm) and width (cm) of the transplanted tumor. Mice were euthanized when tumors reached a volume of ∼2 cm 3 . Tumors were resected and either fixed in 4% formalin for morphological and immunohistochemical analysis or snap-frozen in isopentane/liquid nitrogen for immunofluorescence and stored at −80°C. Statistical analysis of tumor size at day 33 was performed using the Mann-Whitney test and mice survival with the log rank test.

Northern Blot Analysis

Total RNA from tumors and/or cells was prepared using the standard Trizol method (Life Technologies, Inc., Gaithersburg, MD). RNA was analyzed by Northern blot analysis using VEGF and ribosomal protein L28 hybridization probes as described previously (37) .

Southern Blot Analysis

Genomic DNA was isolated from tumors or cells by standard procedures (38) , digested with BamHI, and further analyzed as described previously (39) . Southern blots were hybridized with a 5′ external probe obtained from the digestion of genomic clone λH1 (35) with EcoRV/ClaI. The neo probe was obtained by digesting the pGT1.8IRESβgeo plasmid with BamHI.

Protein Extractions

Tumor tissue (100–200 mg) was homogenized in 500 μl of 50 mm Tris-HCl (pH 7.4); 150 mm NaCl; 10 mm EDTA; 0.25% Triton X-100; 0.1% NP40; the proteinase inhibitors phenylmethylsulfonyl fluoride (1 mm), aprotinin (1 μg/ml), leupeptin (1 μg/ml), and pepstatin (1 μg/ml); and the phosphatase inhibitor sodium vanadate (1 mm). Homogenates were centrifuged (20 min at 16,000 × g), and the supernatant was used for Western blots and VEGF protein analysis.

For PARP detection, tissue protein extracts or cell lysates were diluted in 62.5 mm Tris-HCl (pH 6.8), 6 m urea, 10% glycerol, 2% SDS, 5% β-mercaptoethanol, and 0.00125% bromphenol blue, and then were sonicated (15 s) and heated at 65°C for 15 min before loading. The total protein level in each group was determined by the BCA protein assay (Pierce Chemical, Rockford, IL).

For Western blot analysis of HIF-1α and ARNT, nuclear extracts from ES cells were prepared as described previously (11) except that 0.5% NP40 was used for cell lysis.

Western Blot Analysis

Protein extracts were electrophoresed through SDS-polyacrylamide gels and electrotransferred to nitrocellulose membranes (Schleicher & Schuell, Dassel, Germany) by standard procedures. Membranes were stained with Ponceau S (Sigma) to confirm equal protein loading and transfer. HIF-1α was detected with a chicken anti-HIF-1 polyclonal IgY antibody described previously (40) . Mouse anti-ARNT (2B10) monoclonal antibody was purchased from Affinity Bioreagents (Golden, CO).

PARP antibody (1:10; Oncogene Research Products, Darmstadt, Germany) was used for apoptosis detection. Blots were normalized using β-actin (1:10,000; Pierce). Goat antimouse horseradish peroxidase-coupled secondary antibody (1:1000; Pierce) was applied, followed by chemiluminescence detection (13) .

VEGF Protein Analysis

Immunoreactive VEGF was quantified using a sandwich ELISA (Quantikine M Mouse VEGF Immunoassay kit; R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions.

Immunofluorescence

Cryostat sections (7 μm) were fixed for 10 min in cold acetone, air-dried for 10 min, and blocked with TBS-T (50 mm Tris, 100 mm NaCl, 0.05% Tween 20) containing 20% FCS for 15 min. Sections were incubated with rat anti-PECAM-1 antibody (1:250; PharMingen, San Diego, CA) diluted in 3% BSA-TBS-T and washed in TBS-T. Sections were incubated with a goat antirat Alexa 488 antibody (1:300; Molecular Probes, Leiden, the Netherlands) and 4′,6-diamidino-2-phenylindole (1:1000; Boehringer Mannheim, Mannheim, Germany) in 3% BSA-TBS-T. After sections were washed with TBS-T and mounted in Dabco solution (DAKO, Carpinteria, CA), they were analyzed by fluorescence microscopy at ×25 magnification (described below).

Microscopy and Digital Vessel Length Analysis

The sections were studied by epifluorescence (Polyvar microscope; Reichert Jung, Vienna, Austria). Images were acquired with a VISICAM CCD camera (Visitron, Puchheim, Germany) and processed by Image-Pro Plus v3.0 software (Media Cybernetics, Silver Spring, MD). Slides were double blinded, and 4′,6-diamidino-2-phenylindole staining was used for random choice of the areas to be photographed for PECAM-1 fluorescence to assure no influence of the operator in the final results. Six to 10 nonoverlapping pictures were taken from each section, depending on section size, so that a representative section of the tumor slice was photographed. Pictures were analyzed using the Adobe Photoshop program (Adobe Inc, Mountain View, CA) plug-in “The image processing tool kit” (Reindeer Games, Inc., Asheville, NC) as described (41) .

Morphological Analysis and Immunohistochemistry

Representative parts of all tumors were fixed in buffered 4% formalin, embedded in paraffin, and cut through the plane with the greatest extension. The total area of all sections of each tumor was determined under an axioscope microscope (Zeiss Axioplan; Oberkochen, Germany) with integrated morphometric device (400x magnification). Four distinct groups were defined and analyzed according to their morphological characteristics. Within the total tumor area, each of the components were marked morphometrically, and the relative proportion of each component was measured by adding up all corresponding areas. The percentage of a given tissue within a tumor was calculated by dividing the sum of all areas covered by this tissue by the total area of the slide. Serial sections were stained with a rat antimouse antibody raised against the Ki-67 antigen TEC-3 (DAKO), which detects proliferation by staining the nuclei of dividing cells. The two highly proliferative components, immature neuronal and carcinomatous, were examined separately (×100 magnification) by marking representative areas in all tumors. With the morphometric device, all cells were counted, and the percentage of TEC-3-stained cells in each component was calculated. Statistical differences were determined using the Mann-Whitney test.

Power Doppler Analysis

Precontrast imaging of the entire volume of each tumor was performed according to previously published methods (42 , 43) with a commercially available diagnostic ultrasound unit (ATL 5000, L12-5 transducer; Philips AG, Philips Medical Systems, Zurich, Switzerland). Power Doppler settings were constant for all tumors (81% power gain; medium wall filter; 500-Hz pulse repetition frequency; low velocity flow optimization). The entire tumor was imaged in 1-mm increments along the long axis. Grayscale imaging was performed first to delineate tumor borders; the power Doppler sample volume was then sized and positioned to include the entire tumor and immediate surrounding tissue. Postcontrast power Doppler images were obtained in a similar manner during and immediately after the slow i.v. hand injection of 0.25 ml a 400 mg/ml suspension of monosaccharide microparticles (Levovist; Schering AG, Lucern, Switzerland). Regions of low and high RBC flux and possible areas of necrosis in the tumor were mapped during power Doppler imaging to guide subsequent needle electrode insertion. Five representative power Doppler images each from the pre- and postcontrast series were selected for analysis. On each image, a region of interest was drawn along the tumor boundary, using image processing and analysis software (QWin; Leica Microsystems AG, Zurich, Switzerland). Pre- and postcontrast blood flow parameters for each tumor were determined by averaging the individual quantitated parameters for each of the five images of the series.

pO₂ Measurements

The tumor oxygen tension values were measured polarographically with a sterile needle electrode (Eppendorf, Hamburg, Germany) as described elsewhere (44) . The pO₂ values were recorded automatically while the electrode moved through the tumor tissue in forward steps of 0.7 mm immediately followed by a backward step of 0.4 mm to minimize compression effects. After each backward step, a single pO₂ value was recorded.

RESULTS

Generation of HIF-1α-deficient ES Cells.

The mouse Hif1a is a single-copy gene containing 15 exons (39) . Exon II encodes for the bHLH domain necessary for HIF-1α DNA binding and for dimerization of HIF-1α with ARNT (HIF-1β; Ref. 45 ). The targeting vector was designed to disrupt by homologous recombination the mouse Hif1a gene (36) by replacing exon II with a promotorless IRESβgeo cassette (Fig. 1A) ⇓ . Wild-type (+/+) HM-1 ES cells were electroporated, and clones heterozygous (+/−) for the recombinant allele were selected with G418. Southern blot analysis of genomic DNA digested with BamHI and hybridized with a 5′ external probe (Fig. 1A) ⇓ , showed that 61 of 115 clones were heterozygous for HIF-1α deficiency (data not shown). Hybridization with a neo probe indicated no additional genomic integrations of the targeting vector (data not shown). Heterozygous HIF-1α^+/− (Fig. 1B ⇓ , Lane 2) clone 33 cells were then subjected to increased G418 (4 mg/ml), and three subclones (clones C, L, and M) were selected that showed homozygosity (−/−) for the recombinant allele (Fig. 1B ⇓ , Lane 3). Karyotype analysis of heterozygous HIF-1α^+/− clone 33 and homozygous HIF-1α^−/− clones C, L, and M revealed a normal chromosome constitution (data not shown). To determine the presence or absence of HIF-1α and ARNT at the protein level, ES cells were cultivated for 6 h in either normoxic (21% O₂) or hypoxic (1% O₂) conditions with DFX (260 μm) or CoCl₂ (100 μm). Nuclear extracts from HIF-1α^+/+ and HIF-1α^−/− ES cells were analyzed by Western blot. As shown in Fig. 1C ⇓ , HIF-1α could be detected in HIF-1α^+/+ but not in HIF-1α^−/− ES cells after stimulation by hypoxia, DFX, and CoCl₂. ARNT was slightly increased in nuclear extracts in the presence of all three stimuli in HIF-1α ^+/+, but not in HIF-1α^−/− ES cells. Taken together, these results show that the Hif1a gene was disrupted by homologous recombination, leading to the absence of HIF-1α protein in the three HIF-1α^−/− clones tested.

HIF-1α Confers Growth Advantage to ES Cells Grown in Hypoxia and under Differentiating Conditions.

LIF is necessary to maintain ES cells in an undifferentiated and pluripotent state and is therefore added to culture medium (46) . To assess the role of HIF-1α in cell growth and differentiation, HIF-1α^+/+ and HIF-1α^−/− ES cells were cultivated as monolayers in undifferentiating conditions in the presence of LIF (10³ units/ml) under normoxic or hypoxic conditions (Fig. 2, A and B) ⇓ . As indicated by protein concentrations, the HIF-1α^+/+ cell number was significantly lower than those of the three HIF-1α-deficient clones after 5 days in normoxia (Fig. 2A) ⇓ , whereas under hypoxic conditions the HIF-1α^+/+ and HIF-1α^−/− clone M cell number was higher than those for HIF-1α^−/− clones L and C (Fig. 2B ⇓ ; n = 6).

Withdrawal of LIF induces ES cell differentiation (46) , mimicking the in vivo situation in teratocarcinomas in which ES cells differentiate into all cell types. Thus, ES cells were further cultivated in the absence of LIF under normoxic or hypoxic conditions either as monolayers (Fig. 2, C and D) ⇓ or as EBs (Fig. 2, E and F) ⇓ . In normoxia and in the absence of LIF, the number of HIF-1α^+/+ ES cells was significantly lower than the numbers of HIF-1α^−/− ES clones at day 6 (Fig. 2C) ⇓ . Interestingly, the HIF-1α^+/+ ES cell number was higher than those of the HIF-1α^−/− ES cells under hypoxic and differentiating conditions (Fig. 2D) ⇓ . These results were further confirmed by differentiating ES cells into three-dimensional aggregates, called EBs (47) . Fig. 2E ⇓ shows that under normoxic conditions, HIF-1α^+/+ EBs grew significantly slower than HIF-1α^−/− EBs, whereas under hypoxic conditions the opposite occurred (Fig. 2F) ⇓ . These in vitro results suggest that lack of HIF-1α under differentiating conditions confers a growth advantage to ES cells or EBs in normoxia. In hypoxia, however, HIF-1α-deficient ES cells or EBs grew significantly slower than their HIF-1α^+/+ counterparts.

Western blot analysis for the nuclear enzyme PARP, a hallmark of apoptosis, was performed on total cellular lysates from HIF-1α^+/+ and HIF-1α^−/− ES cells or EBs exposed to normoxia or hypoxia (1% O₂) for 5 (ES cells) and 6 (EBs) days. Similar PARP cleavage was detected in both HIF-1α^+/+ and HIF-1α^−/− ES cells at day 5 and in EBs at day 6 (data not shown), suggesting that the decreased growth of the HIF-1α-deficient ES cells or EBs under hypoxic conditions is not attributable to apoptosis.

Absence of HIF-1α Leads to a Decrease in Tumor Mass and Growth.

ES cells develop into teratocarcinomas when injected into athymic mice. To determine the role of HIF-1α in solid tumor formation, we injected 5 × 10⁶ HIF-1α^+/+ and HIF-1α^−/− ES cells (clones L or M) s.c. into BALB/c nu/nu mice (n = 10/ES cell line). Because tumors are not only constituted of neoplastic cells, but also of, e.g., endothelial cells or infiltrating cells, HIF-1α^+/+ ES cells were mixed with HIF-1α^−/− ES cells at a ratio of 1:10 and 1:100 (n = 10/ES cell mixture). Kaplan-Meier analysis was performed to estimate the probability that the HIF-1α^−/− tumors would reach 2 cm 3 after a certain period of time (Fig. 3A) ⇓ . The median survival time for the mice bearing HIF-1α^+/+, 1:10 mixed, and 1:100 mixed tumors was 46, 43, and 43 days, respectively. At day 49, when all mice were sacrificed, 70% of the HIF-1α^−/− tumor-bearing mice had tumor volumes <2 cm 3 . Tumor volumes were significantly larger in HIF-1α^+/+, 1:10 mixed, and 1:100 mixed tumors compared with HIF-1α^−/− teratocarcinomas at day 33, when all mice were still alive (Fig. 3B) ⇓ . No significant difference was measured among the growth rates of the HIF-1α^+/+, 1:10 mixed, and 1:100 mixed tumors during the course of the experiment (Kruskal-Wallis test, P > 0.05). Furthermore, HIF-1α^+/+, 1:10 mixed, and 1:100 mixed tumors grew 4.0, 7.0, and 3.6 times faster than the HIF-1α^−/− tumors, respectively. Interestingly, when 1:10 or 1:100 mixtures of wild-type Hepa1 and ARNT mutant Hepa1C4 cells were injected into CBA nu/nu mice, Hepa1 wild-type and mixed tumors grew faster than the Hepa1C4 tumors (data not shown). These results demonstrate that the absence of HIF-1α significantly impairs tumor growth and that tumor growth can be rescued by adding wild-type ES cells to HIF-1α^−/− ES cells at a ratio of only 1:10 or 1:100.

Lack of Difference in Blood Flow and Vessel Density among HIF-1α^+/+, HIF-1α^−/−, and Mixed Tumors.

To determine whether impaired vascularization could be responsible for the decrease in growth of the HIF-1α^−/− tumors, we first assessed vascularity noninvasively by power Doppler ultrasound analysis before and after i.v. ultrasound contrast medium injections. Three blood flow parameters were measured: MCL, a measure of mean RBC flux or the mean number of moving RBCs; FA, which represents the average tumor cross-sectional area containing blood flow; and CWFA, the product of MCL and FA, which represents overall blood flow in the tumor (48) . No significant difference was observed in either the CWFA or FA of the HIF-1α^+/+ versus HIF-1α^−/− tumors before (data not shown) or after contrast injection (Fig. 4A) ⇓ . These results were further confirmed by immunofluorescence using an antibody against the endothelial cell marker PECAM-1 (CD31; Fig. 4B ⇓ ) and by imaging the blood vessels by fluorescent microscopy (Fig. 4, C and D) ⇓ . Interestingly, quantification of vessel density by computer-assisted image analysis did not show any significant difference in the microvessel density among the HIF-1α^+/+, HIF-1α^−/−, and mixed tumors (Fig. 4B) ⇓ . These results suggest that lack of neovascularization is not responsible for a decrease in HIF-1α^−/− tumor growth.

VEGF mRNA Expression Does Not Reflect VEGF Protein Synthesis.

VEGF was examined at the mRNA and protein levels in HIF-1α^+/+, HIF-1α^−/−, and mixed tumors. Quantification of the relative mRNA levels by Northern blotting (Fig. 5) ⇓ revealed that VEGF is expressed in HIF-1α-deficient and mixed tumors at significantly lower amounts than in HIF-1α^+/+ tumors. There was no significant difference between the HIF-1α^−/− and mixed tumors (Kruskal-Wallis test, P > 0.05). Because the blood flow and vessel density results contradicted the VEGF mRNA expression, we determined whether mRNA levels reflected the VEGF protein levels. Thus, proteins were extracted from all tumors, and the VEGF protein concentration was measured by ELISA. As shown in Fig. 5C ⇓ , there was no significant difference in the amount of VEGF protein among the HIF-1α^+/+, HIF-1α^−/−, and mixed tumors.

Similar Rate of Apoptosis Is Detected in HIF-1α^+/+, HIF-1α^−/−, and Mixed Tumors.

Because neovascularization did not seem to be the limiting factor in the growth reduction of HIF-1α^−/− tumors, we investigated whether limited growth was attributable to increased apoptosis in the HIF-1α^−/− compared with the HIF-1α^+/+ and mixed tumors. For this we analyzed expression of the nuclear enzyme PARP, a hallmark for apoptosis. Proteins were extracted from all tumors, and Western blot analysis was performed using the anti-PARP Ab-2 antibody (49) . The uncleaved form of PARP (115 kDa) and the cleavage product (85–90 kDa) were detected in HIF-1α^+/+, HIF-1α^−/−, and mixed tumors (Fig. 6A) ⇓ .

One mechanism by which hypoxia induces apoptosis is p53 dependent, and it has been reported to be activated at 0.02%, but not at 0.2% O₂ (50) . To investigate whether tumor apoptosis was caused by low pO₂, we measured the degree of hypoxia by polarography in HIF-1α^+/+ and HIF-1α^−/− tumors. As shown in Fig. 6B ⇓ , there was no significant difference in the median pO₂ between HIF-1α^+/+ and HIF-1α^−/− tumors. Similar results were obtained with pO₂ measurements in wild-type Hepa1 and ARNT mutant Hepa1C4 tumors. 4 Furthermore, Northern blots prepared with total RNA extracted from teratocarcinomas were probed for p21, a p53 target gene. No difference between the two genotypes was found (data not shown), suggesting that hypoxia-inducible p53-dependent apoptosis does not occur in these tumors.

HIF-1α^+/+ Cells Do Not Overgrow HIF-1α^−/− Cells in the Mixed Tumors.

Mixed tumors show increased growth rates and tumor volumes compared with pure HIF-1α^−/− teratocarcinomas. One possible explanation would be that HIF-1α^+/+ cells overgrow the HIF-1α^−/− cells. To estimate the relative contribution to the tumor mass, genomic DNA was extracted from all tumors and analyzed by Southern blotting. As shown in Fig. 7 ⇓ , the HIF-1α wild-type allele is present in HIF-1α^+/+ tumors (Lanes 1–4), whereas the recombinant allele (Lanes 5–8) is present in HIF-1α^−/− tumors. However, in all mixed tumors tested (Fig. 7 ⇓ , Lanes 9–16), the wild-type allele was faintly detected only in one of ten 1:100 tumors (Fig. 7 ⇓ , Lanes 13–16) and in none of the 1:10 tumors (Fig. 7 ⇓ , Lanes 9–12). HIF-1α^+/+ cells in the 1:10 and 1:100 tumors contributed <50% to the tumor mass compared with the control representing an in vitro 1:1 mixture of HIF-1α^+/+ and HIF-1α^−/− genomic DNA (Fig. 7 ⇓ , Lane 17). These data demonstrate that HIF-1α^+/+ cells did not overgrow HIF-1α^−/− cells. In support of these data, Hepa1 wild-type cells irradiated with a dose of 25 Gy to prevent subsequent proliferation were able to rescue the growth of ARNT mutant Hepa1C4 tumors at the same extent as in untreated Hepa1 wild-type cells. 4

Differences in the Cellular Composition between HIF-1α^+/+ and HIF-1α^−/− Tumors.

The cellular composition of the tumors was analyzed by histopathology and immunohistochemistry. The teratocarcinomas derived from HIF-1α^+/+, HIF-1α^−/−, and mixed ES cells were composed of mature epithelial structures (epidermis-like, skin appendages, glands of different type), mature mesenchymal tissue (connective tissue, muscle, cartilage, bone), mature neuronal tissue (round nuclei with abundant fibrillary matrix), immature neuronal tissue (large mitotic nuclei, scanty cytoplasm; cells forming rosettes), and carcinomatous tissue (large mitotic nuclei, scanty eosinophilic cytoplasm; undifferentiated cells forming sheets). HIF-1α^+/+ tumors contained significantly more mature neuronal tissues compared with the HIF-1α^−/− tumors, whereas there was no significant difference in the epithelial, immature neuronal and carcinomatous components (Fig. 8A) ⇓ . In contrast, HIF-1α^−/− tumors contained significantly more mature mesenchymal tissue. Interestingly, examination of the 1:10 and 1:100 mixed tumors showed a relative proportion of contributing tissue components similar to the proportion in the pure HIF-1α^−/− tumors, whereas the growth was equivalent to that of HIF-1α^+/+ tumors (Fig. 8B) ⇓ .

To determine the proliferating components of the tumors, we performed immunohistochemistry with a rat antibody derived against the mouse Ki-67 antigen (TEC-3). Two highly proliferative components were stained, representing immature neuronal and carcinomatous tissue. These components were examined separately by marking representative areas in all tumors. Using a morphometric device, we counted all cells and calculated the proportion of TEC-3 stained cells. The immature neuronal and carcinoma compartments in the HIF-1α^+/+ tumors had a proliferative activity 2.5 and 2.4 times, respectively, that of HIF-1α^−/− tumors. The 1:10 mixed tumors showed an increased proliferation rate 2.0 times the rate of the immature neuronal tissue and 1.8 times the rate of the carcinoma tissue in the HIF-1α^−/− tumors. However, 1:100 mixed tumors did not show a significant increase in the proliferation rate of the cells of these components (data not shown).

DISCUSSION

Clinical studies using oxygen-sensitive electrodes provided evidence that human tumors contain regions of low pO₂. Tumor hypoxia correlates with poor prognosis after treatment (3 , 51) . Recent efforts concentrated on this unique feature of solid tumors to develop antitumor strategies, including the production of drugs that are toxic under hypoxic conditions or the generation of hypoxia-inducible vectors containing concatamerized HRE sites. These vectors activate toxin production under hypoxic conditions (for review, see Ref. 6 ). However, the role of HIF-1α in tumorigenesis is still controversial. Our study provides new information concerning the implication of HIF-1α in tumor growth, vasculogenesis, apoptosis, and differentiation. We generated teratocarcinomas in nude mice by injecting HIF-1α^+/+ and HIF-1α^−/− ES cells. In agreement with previous publications on HIF-1α-deficient ES cells (52) and on transformed mouse embryonic fibroblasts (52) , our results show that HIF-1α^+/+ tumors grow significantly faster than HIF-1α^−/− tumors. Similarly, and as reported by others (53) , tumor growth of HIF-1β-negative Hepa1C4 cells was retarded compared with the wild-type Hepa1 tumors (data not shown). Furthermore, Kung et al. (28) demonstrated that blockage of HIF-1 interaction with its coactivators p300 and cAMP-responsive element binding protein diminished tumor growth. However, these results contradict other published reports on the role of HIF-1α in tumorigenesis (29 , 54) . This discrepancy may be attributable to the chosen “wild-type” control, which was a recombinant HIF-1α^+/+ ES cell line harboring a randomly integrated HIF-1α gene-targeted vector called rHIF-1α^+/+, which survived to high G418 selection. Depending on the integration site of the vector, it may have disrupted a gene involved in cell growth and apoptosis, leading to diminished growth of the wild-type tumor (29) .

Tumors that reach diameters of several millimeters develop microregions of hypoxia and therefore require new blood vessel formation to supply cells with oxygen and nutrients. HIF-1α is involved in neovascularization by inducing the angiogenic growth factor VEGF under low pO₂. Measurement of blood flow by power Doppler ultrasound and estimation of blood vessel density by immunofluorescence with anti-PECAM-1 antibodies showed no difference between HIF-1α^+/+ and HIF-1α^−/− tumors. This blood vessel density result is consistent with a report by Ryan et al. (52) , in which the authors induced tumorigenesis in nude mice with HIF-1α wild-type or HIF-1α-deficient immortalized and transformed mouse embryonic fibroblasts. We previously demonstrated in vitro that hypoxic induction of VEGF mRNA, but not basal expression, was diminished in HIF-1α-deficient ES cells (55) . In vivo, we showed that VEGF mRNA is decreased but not suppressed in HIF-1α^−/− tumors, corroborating with results obtained by others (27 , 29 , 52) . Interestingly, and in contrast to VEGF mRNA expression, we did not find any significant difference in VEGF protein levels. These findings can be explained either by the fact that, in hypoxia, VEGF mRNA accumulates by the prolongation of its half-life through posttranscriptional regulatory events mediated by specific cis-sequences in the 3′ untranslated region (56 , 57) or by the infiltration of VEGF-expressing host fibroblasts into the tumor (58) . Moreover, as reported by Iyer et al. (59) and Kotch et al. (60) in tissue culture cells, VEGF mRNA expression is induced by glucose deprivation independently of HIF-1α, therefore providing a mechanism by which VEGF mRNA expression is increased in HIF-1α^−/− embryos. Our results strongly indicate that, in tumors lacking HIF-1α, VEGF is still produced in amounts that are sufficient to induce angiogenesis and that the negative effect on HIF-1α^−/− tumor growth is not caused by deficient vascularization.

Carmeliet et al. (29) suggested that HIF-1α-dependent growth arrest/apoptosis under hypoxic conditions was through p53-mediated control of growth arrest/apoptosis. However, we did not find any in vivo p53-responsive p21 induction by Northern blot or Western blot analyses. Furthermore, PARP cleavage, a hallmark of apoptosis, was detected in HIF-1α^+/+, HIF-1α^−/−, and mixed teratocarcinomas to a similar extent. Hence, apoptosis is probably not implicated in the diminished growth rates and tumor volumes of the HIF-1α-deficient teratocarcinomas.

We added HIF-1α^+/+ cells to HIF-1α^−/− ES cells to determine whether HIF-1α^+/+ cells have a growth advantage in this competitive situation and whether HIF-1α^+/+ cells can rescue the growth alteration of HIF-1α-deficient tumors. Using mixtures of HIF-1α^+/+ cells to HIF-1α-deficient cells at proportions as low as 1:10 and 1:100, we showed that the tumors will behave like wild-type teratocarcinomas with regard to growth rate and tumor volume. Analysis of genomic DNA showed that HIF-1α^+/+ cells did not overgrow HIF-1α^−/− cells, and morphometric analyses of the tumors demonstrated that the cellular composition of the mixed tumors was the same as that of the HIF-1 α^−/− tumors. The presence of HIF-1α^+/+ cells in the mixed tumors did not change the differentiation pattern of the HIF-1α^−/− cells predominating in these tumors, suggesting that differentiation depends on the cellular genotype and cannot be modified by secreted factors induced by hypoxia. In contrast to differentiation, tumor growth was highly influenced when HIF-1α^+/+ cells were mixed with HIF-1 α^−/− cells, suggesting that mixed tumor growth is mediated by soluble factor(s) secreted by HIF-1α^+/+ cells. Furthermore, the fact that the influence of HIF-1α in differentiation is independent of its effect on tumor growth provides evidence that results on growth behavior obtained from teratocarcinomas can be transferred to other types of tumors. We confirmed this notion by mixing wild-type Hepa1 with ARNT-mutated Hepa1C4 cells. Thus, these differences in cellular components and differentiation of teratocarcinomas do not influence tumor growth. Our results imply that the presence of HIF-1α^+/+ ES cells in the tumor provides a HIF-1α-inducible growth factor, which allows the HIF-1α-deficient cells to proliferate. Successful development of an anti-HIF-1α tumor therapy requires targeting of hypoxic or genetically transformed HIF-1α-overexpressing cells, but our data suggest that a minor population of surviving HIF-1α-positive cells might rescue tumor growth even in the absence of HIF-1α, which would drive a selection process that results in tumor resistance to anti-HIF-1 therapy.